Few changes to EU data law matter more to teams that collect web data for AI than the Digital Omnibus. Proposed by the European Commission a year ago, it would rewrite parts of the GDPR, including the rules on AI training, pseudonymised data and cookies. In October 2026, EU governments agreed their own version, and it differs from the Commission’s in ways that matter. This guide covers where the file stands, what each side wants, and what applies to AI and machine learning data collection today. It is general information, not legal advice.
Where it stands
| Date | Step |
|---|---|
| 19 November 2025 | Commission proposes the Digital Omnibus, including COM(2025) 837 on the GDPR and cookies |
| 10 February 2026 | European data protection regulators publish a joint opinion criticising key parts |
| 7 July 2026 | Regulators adopt draft guidelines on web scraping for generative AI, for public consultation |
| 21 October 2026 | EU governments in the Council agree their negotiating mandate |
| Not yet | European Parliament adopts its position, then the three institutions negotiate |
Nothing in the Omnibus applies yet. The current GDPR still governs every scraping and AI training project in the EU.
What each side wants
The main changes for web data teams, as proposed by the Commission and as amended by EU governments:
| Topic | Commission proposal | Council mandate |
|---|---|---|
| Definition of personal data | Data is not personal for an entity that cannot identify the person | Deleted: the current definition stays |
| Pseudonymised data | No separate rule | Not personal data for a third party unable to identify the person |
| AI training | Legitimate interest can be used, with an “unconditional right to object” | Legitimate interest can be used, with the normal right to object and a case-by-case test |
| Sensitive data in training sets | Residual sensitive data allowed if controllers avoid and remove it | Kept for incidental residual data, which must be erased “without undue delay” |
| Cookies | Moved into the GDPR, with browser-based consent signals | Kept in the ePrivacy rules, with six consent exemptions and a single-click refusal |
| Data breaches | Notify only high-risk breaches, within 96 hours | Kept |
The AI training question
For teams building training sets from the web, the AI provision is the heart of it. The Commission proposed a new article saying that processing for the development and operation of AI may be based on legitimate interest, one of the six legal bases in the GDPR. EU governments kept the idea but added a brake. Their text says “the development and operation of an AI system or AI model does not in itself constitute a legitimate interest”, and that the legal basis “is to be assessed on a case-by-case basis, taking into account the underlying purpose of that processing.”
In practice, this changes less than the headlines suggest. Regulators already accept that legitimate interest can cover AI development. Their joint opinion on the Omnibus says the European Data Protection Board “has already explicitly confirmed this in its Opinion 28/2024”, and that “it is not necessary to add a specific provision to the GDPR on this point”. The balancing test, the safeguards and the duty to respect objections stay either way.
The pseudonymisation question
The second change with real reach concerns identifiability. The Commission wanted data to count as personal only for those who can identify the person behind it, following a 2025 ruling of the EU Court of Justice. Regulators pushed back hard: they “strongly urge the co-legislators to not adopt the proposed changes to the definition of personal data.”
EU governments struck the change and wrote a narrower rule instead: “Pseudonymised data shall not be considered personal data for third party if that third party is unable to identify the natural person to whom the data relates.” For data teams, that matters when sharing or buying datasets: properly pseudonymised data handed to a party that cannot re-identify it could fall outside the GDPR for that party.
Cookies, briefly
The Council also reversed the Commission’s plan to move cookie rules into the GDPR. Its version keeps them in the ePrivacy Directive and adds consent exemptions, including for first-party audience measurement and security. When consent is required, people must be able to refuse “with a single-click button or equivalent means”, and a site may not ask again for the same purpose “for a period of at least four months”. Our look at consent walls by country shows how differently sites handle consent today.
What applies today
Until the Omnibus is adopted, three documents set the practical rules for scraping personal data in the EU:
- EDPB Opinion 28/2024 on AI models, December 2024, which accepts legitimate interest with a three-step test and lists safeguards for scraping, such as excluding websites that object to it.
- EDPB Guidelines 03/2026 on web scraping in the context of generative AI, adopted for public consultation on 7 July 2026. They say scrapers should “exclude from the collection websites which clearly oppose the scraping of their content”, through measures such as “robots.txt or ai.txt files, or CAPTCHA”.
- National guidance, such as France’s CNIL, which says “The collection of personal data available online through web scraping is generally based on legitimate interest”, with additional measures to limit the impact on people.
The guidelines are still a draft, but the direction is clear: technical opt-out signals carry legal weight. Our guide to robots.txt and AI opt-out signals covers the signals in use.
What data teams should do now
- Plan for today’s GDPR. The Omnibus may take a year or more to finish. Build on legitimate interest with a documented assessment, as regulators already expect.
- Honour opt-outs. Respect robots.txt, ai.txt and similar signals for AI collection, and keep a record that you did.
- Filter sensitive data. Both versions of the Omnibus expect sensitive data to be avoided and removed. Build filters now.
- Pseudonymise early. If the Council’s rule survives, well-pseudonymised data shared with others could be treated more simply. It is good practice anyway.
- Collect openly. Use residential proxies to see public pages as a local visitor would, not to get past a site that has said no. Our guide to ethical residential proxies for AI data collection sets out where we draw that line.
Outside the EU, the rules are moving in their own directions. Our explainer on the UK Data (Use and Access) Act covers the British approach.
The reactions
The proposal split opinion from the start. Privacy group noyb called it “the biggest attack on European’s digital rights in years”. After the Council deal, the digital rights network EDRi said: “The result is a deal that weakens important safeguards and makes redress harder.” Industry groups such as DIGITALEUROPE have welcomed the aim of clarifying how the GDPR applies to innovation. The final text will depend on the Parliament, which is still working on its position.
The bottom line
The Digital Omnibus would confirm that AI training can rely on legitimate interest, but EU governments have made clear it is no free pass, kept the current definition of personal data, and added a narrower rule for pseudonymised data. None of it applies yet. For now, the current GDPR and the regulators’ draft scraping guidelines set the rules, and both point the same way: document your basis, filter sensitive data and respect sites that opt out.
Sources and references
- European Commission, Digital Omnibus proposal on data protection and privacy, COM(2025) 837 (19 November 2025): eur-lex.europa.eu
- European Commission, press release IP/25/2718 (19 November 2025): ec.europa.eu
- Council of the EU, mandate for negotiations with the European Parliament, ST 14756/26 (October 2026): data.consilium.europa.eu
- European Parliament, Legislative Observatory, procedure 2025/0360(COD): europarl.europa.eu
- EDPB and EDPS, Joint Opinion 2/2026 on the Digital Omnibus (10 February 2026): edpb.europa.eu
- EDPB, Opinion 28/2024 on AI models (17 December 2024): edpb.europa.eu
- EDPB, Guidelines 03/2026 on web scraping in the context of generative AI, version for consultation (7 July 2026): edpb.europa.eu
- CNIL, focus sheet on web scraping and legitimate interest: cnil.fr
- noyb, statement on the Digital Omnibus (19 November 2025): noyb.eu
- EDRi, statement on the Council position (November 2026): edri.org